Abstract

In recent years, point sampled geometry models have received growing attention in computer graphics. Compared with polygon based models, the quality of rendered images for point sampled geometry is much lower, so far. This is primarily due to the discrete form of representing a 3D surface just by sample points. Clearly more effort is needed in rendering techniques for point sampled models. In this thesis, we describe the results of our efforts towards producing high quality images for point based geometry. We adapt a number of advanced mesh rendering techniques for discrete surfaces represented as point clouds. These techniques include self shadowing effects using ambient occlusion, diffuse lighting estimation using spherical harmonic representations of irradiance environment maps and specular lighting estimation by casting a reflected ray at each point into a pre-blurred version of the environment map. Ambient occlusion computations are improved by using octree hierarchy based on feature analysis of the point cloud data. Other lighting calculations are performed directly on the Graphics Processing Unit (GPU) using vertex and fragment shaders. We improve the overall rendering quality and speed by combining pre-computed ambient occlusion results with environment lighting and procedural textures directly on the GPU. The adapted techniques have been implemented as a plug in for the Pointshop3D public domain software system.